How does this skill detect prompt injection?

It utilizes a robust set of regular expression patterns to identify common injection tactics like instruction overrides, role-change attempts, and special token manipulation.

Is this skill compatible with any LLM?

Yes, these security patterns are provider-agnostic and act as a middleware layer that works with Claude, GPT, or any other Large Language Model.

What content categories are filtered in the output?

The filter screens for several unsafe categories including violence, hate speech, sexual content, self-harm, and illegal activities.

Can I recover redacted PII from the output?

Yes, the skill provides a PIIRedactor class that uses a mapping system to replace sensitive data with unique tokens, allowing you to unredact information once the LLM finishes processing.

AI Security & LLM Guardrails

Name: AI Security & LLM Guardrails
Author: ricardoroche

byricardoroche

セキュリティとテスト

Secures LLM applications by implementing prompt injection detection, PII redaction, and output content filtering.

概要

This Claude Code Skill provides a comprehensive suite of security patterns designed to protect AI-powered applications from modern threats. It automates the implementation of critical safety layers, including heuristic-based prompt injection detection, sensitive PII redaction with secure token mapping, and content moderation filters for LLM outputs. By utilizing these standardized patterns, developers can prevent jailbreaks, ensure data privacy, and maintain strict content safety standards throughout the entire LLM lifecycle from input sanitization to response validation.

主な機能

Secure prompt construction using standardized system/user delimiters.
Multi-category output filtering for violence, hate speech, and illegal content.
0 GitHub stars
Advanced prompt injection detection for identifying instruction overrides and role-play attacks.
Automated PII redaction with unique tokenization and restoration capabilities.
Integrated rate limiting and abuse prevention mechanisms.

ユースケース

Protecting enterprise chatbots from jailbreaks and system prompt leakage.
Sanitizing sensitive user data before transmission to third-party LLM providers.
Implementing content safety guardrails for public-facing AI applications.

概要

主な機能

Secure prompt construction using standardized system/user delimiters.
Multi-category output filtering for violence, hate speech, and illegal content.
0 GitHub stars
Advanced prompt injection detection for identifying instruction overrides and role-play attacks.
Automated PII redaction with unique tokenization and restoration capabilities.
Integrated rate limiting and abuse prevention mechanisms.

ユースケース

Protecting enterprise chatbots from jailbreaks and system prompt leakage.
Sanitizing sensitive user data before transmission to third-party LLM providers.
Implementing content safety guardrails for public-facing AI applications.